About The Dataset

I scraped data from the ATP World Tour website from 1980 through 2018. If there’s enough interest, I’ll make another article detailing how to scrape the site. While player name and associated rank were completely non-missing, there were a substantial amount of missing values in the variable listing the type of backhand a player hits:

p1 <- clean %>% 
  ggplot(aes(year, fill = is.na(one_hander))) +
  geom_density(alpha = .6) +
  labs(title = "Distribution of Missing Backhand Type by Year",
       fill = "Missing?",
       x = "")

ggplotly(p1)

The graph above shows that the vast majority of the missings were for the earlier years, which we might expect.

Perhaps more problematic, however, is that the missing backhand information is correlated with rank; lower ranked players were more likely to contain missing backhand types:

p2 <- clean %>% 
  mutate(Ranking = parse_number(Ranking)) %>% 
  group_by(Ranking) %>% 
  summarize(missing_count = sum(is.na(one_hander))) %>% 
  ggplot(aes(Ranking, missing_count)) +
  geom_line(size = 1.7) +
  expand_limits(y = 0) +
  scale_x_continuous(breaks = c(1,10,20,30,40,50,60,70,80,90,100)) +
  labs(title = "Sum of Missing Backhand Counts By Ranking", y = "")

ggplotly(p2)

In total, 610 unique players were missing backhand types. This isn’t terribly hard to hard code, and I’ll leave that for someone else to do should they want to collaborate (email me!).

Despite these shortcomings, I filtered for dates from 2000 onward and pushed on. Note that the confidence intervals in the graph below are almost invisible due to the abundance of data points in a given year.

ONE HANDED BACKHAND SHARE BY YEAR

p3 <- clean %>% 
  filter(year >= 2000) %>% 
  group_by(date) %>% 
  summarize(share = mean(one_hander, na.rm = TRUE)) %>% 
  ggplot(aes(date, share)) +
  geom_point(size = .6) +
  geom_smooth(size = 1.7) +
  scale_y_continuous(labels = percent_format()) +
  theme(panel.grid.major = element_blank(),
        panel.grid.minor = element_blank()) +
  labs(x = "", y = "", color = "", title = "Share of One-Handers by Year, 95% CI's Shown")

ggplotly(p3)

Above we see that the share of tennis players in the ATP Top 100 hitting one-handed backhands is undoubtedly in steady decline: the percentage has plummeted from around 42% in 2000 to just 18% in 2018. However, this style of swing is often referred to as an “artisan” shot, reserved for the best of the best. As such, I wanted to examine trends separately for the top 50 vs. 51 - 100.

p4 <- clean %>% 
  filter(year >= 2000) %>% 
  group_by(year, rank_cohort) %>% 
  summarize(share = mean(one_hander, na.rm = TRUE)) %>% 
  ggplot(aes(year, share, color = as.factor(rank_cohort))) +
  geom_smooth(size = 1.7) +
  theme(panel.grid.major = element_blank(),
        panel.grid.minor = element_blank()) +
  scale_y_percent() +
  labs(x = "", y = "", color = "", title = "Percent of One Handed Backhands in ATP Top 100, 95% CI's Shown")

ggplotly(p4)